Skip to main content
2023 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

    Device Technologies and Biomedical Robotics

    A Machine Learning and Hardware Solution for Early Risk Detection of Post-Thrombotic Syndrome

    (K-422) A Machine Learning and Hardware Solution for Early Risk Detection of Post-Thrombotic Syndrome

    Saturday, October 14, 2023
    9:30 AM – 10:30 AM PDT
    Location: Exhibit Hall - Row K - Poster # 422

      Presenting Author(s)

    • JL

      Jason Lee

      Undergraduate Research Engineer
      University of Pittsburgh
      Santa Clara, California, United States

      • Co-Author(s)

    • CD

      Cyrus Darvish

      Research Engineer
      University of Pittsburgh, United States

    • PG

      Pete Gueldner

      Graduate Research Student
      University of Pittsburgh, United States

    • DV

      David Vorp

      Senior Associate Dean for Research and Facilities
      University of Pittsburgh, United States

      • Last Author(s)

    • TC

      Timothy Chung

      Research Assistant Professor
      University of Pittsburgh, United States

    Introduction::
    Deep Vein Thrombosis (DVT) is a vascular disease characterized by the formation of blood clots in the veins of the arm, lower leg, or thigh. There are around 900,000 people affected by this condition with nearly 100,000 deaths each year in the US from resulting complications such as pulmonary embolism1. Approximately 50% of DVT cases lead to a more serious condition called post-thrombotic syndrome (PTS)2 associated with serious symptoms such as chronic inflammation, chronic pain, and even permanent disability2. With treatments costing as much as $20,000 per patient coupled with a nearly 30% chance of blood clot recurrence, the US spends $10 billion annually on DVT and related diseases1. Although there are available treatments such as blood-thinners for DVT, there are currently no methods to predict the transition of DVT to PTS. Patients are often only diagnosed with PTS after the onset of major symptoms resulting in the reduced efficacy of possible treatments. It is hypothesized that there is a significant difference in venous wall properties such as compliance, the ability of a vein to change diameter in response to changes in pressure,

    in people with and without PTS. As such, this study aims to design a hardware and machine learning solution to provide accurate and efficient early risk prediction of PTS through the measurement of vein compliance in patients.


    Materials and Methods::

    The existing prototype utilizes arrays of pressure sensors and Arduino Nano microcontroller boards (Arduino, Somerville, MA, USA) to read vein compliance. A total of nine pressure sensors were configured into three separate arrays, each comprised of three sensors. Each array was assigned an individual Arduino Nano v3.0 board connected to a computer through mini-USB connection. The sensor arrays were then embedded in a custom PDMS mold to separate the electrical components from the sensors. The connection between each sensor and the corresponding Arduino Nano of its array is established at an analog pin through a signal pin on an amplifier (Figure 1A). A power distribution block connected to a constant 5-volts power brick supplies constant voltage for the amplifiers. Recordings from the sensors are facilitated by a Python script that utilizes the multiprocessing and pySerial libraries to interface with the Arduinos and perform concurrent data collection across all three Arduinos. A supporting C++ script for supplying sensor values to requests from the Python script was written in Arduino IDE and uploaded to each Arduino board. The analog values directly recorded from the Arduinos are converted to a voltage value using equation 1.




     




                                                 eq. 1




     




    The sensor arrays and linked electrical components were attached onto an inflatable cuff. They were strategically positioned such that the sensor arrays were facing the cuff’s interior, enabling them to record from anterior and lateral sides of a patient’s leg. The electronics and Arduino boards were affixed to the exterior of the cuff (Figure 1B).




    Results, Conclusions, and Discussions::

    Concurrent and rapid data acquisition across all three Arduinos was an important design consideration for getting the most accurate representation of vein compliance in real time. Initially, a single Arduino board and nine analog ports were used to read the sensors. However, high latency between successive readings was observed due to the serial communication limitation of Arduino boards. The low latency of our new set-up provides an improved solution to measure vein compliance in real-time. Simultaneous data collection from the three Arduinos occurred nearly in sync with each same positioned sensor in an array (top, middle, bottom) being recorded at the same time across all boards within the millisecond. Within an individual array, the serial connection of the sensors resulted in about 2-4 milliseconds in latency between each sensor. Each cycle, or the time it took to read a sensor again after a having previously read it, was around 10-11 milliseconds in latency. In total, the Arduinos recorded approximately 12±2 complete cycles every 1/10th of a second. A trial reading was conducted by placing the cuff around a test subject’s upper thigh and inflating it to 240mmHg. The resulting plot of sensor values showed the real-time recorded values across all nine sensors starting at the max pressure and as the pressure was slowly released (Figure 1C).


    This study has laid the groundwork for an innovative solution to predict the transition from DVT to PTS. With a functional prototype, our primary focus going forward will be on data acquisition and data refinement through post-processing methods. Our data will be obtained from a diverse set of patient profiles, encompassing healthy patients and patients diagnosed with DVT and PTS. Upon completion of data acquisition, our efforts will pivot towards the training of a machine learning model to predict the risk of PTS given the measured vein compliance of a patient. The sensitivity and specificity of this model to various patient factors will be tested and fine-tuned. Ultimately, this study, while in its initial stages, carries the potential to significantly improve patient outcomes by providing early and accurate prediction of PTS.



    Acknowledgements (Optional): :
    We would like to acknowledge the University of Pittsburgh Honors College Brackenridge Fellowship for supporting this body of work for Jason Lee.


    References (Optional): :

    [1] “Impact of Blood Clots on the United States”. Centers for Disease Control and Prevention (CDC). June 09, 2022. https://www.cdc.gov/ncbddd/dvt/infographic-impact.html




    [2] Susan R. Kahn; “The post-thrombotic syndrome.” Hematology Am Soc Hematol Educ Program 2016; 2016 (1): 413–418. doi: 10.1182/asheducation-2016.1.413